Skip to content

DOCKER_README.md

Latest commit

 

History

History
98 lines (73 loc) · 3.23 KB

DOCKER_README.md

File metadata and controls

98 lines (73 loc) · 3.23 KB

Docker Images

OpenWeights uses two separate Docker images:

1. Worker Image (nielsrolf/ow-default)

The full worker image includes all GPU dependencies (PyTorch, Unsloth, vLLM) for running ML jobs.

Building and pushing worker images

# Get the current version from the codebase
VERSION=$(python -c "from openweights.client.jobs import Jobs; print(Jobs.base_image.split(':')[-1])")

# Step 1: Build locally for ARM64 (on your Mac)
docker buildx build \
  --platform linux/arm64 \
  -t nielsrolf/ow-default:$VERSION \
  --load .

# Step 2: Build and push AMD64 to Docker Hub
docker buildx build \
  --platform linux/amd64 \
  -t nielsrolf/ow-default:$VERSION \
  --push .

Running worker image locally

# Get the current version
VERSION=$(python -c "from openweights.client.jobs import Jobs; print(Jobs.base_image.split(':')[-1])")

docker run --rm --env-file .env -ti nielsrolf/ow-default:$VERSION /bin/bash

2. Cluster/Dashboard Image (nielsrolf/ow-cluster)

A lightweight image for running the cluster manager and/or dashboard backend. Does not include GPU dependencies.

Building and pushing cluster images

Important: Build the frontend first before building the Docker image:

# Get the current version from the codebase
VERSION=$(python -c "from openweights.client.jobs import Jobs; print(Jobs.base_image.split(':')[-1])")

# Step 1: Build the frontend (run from repository root)
cd openweights/dashboard/frontend
npm install
npm run build
cd ../../..

# Step 2: Build locally for ARM64 (on your Mac)
docker buildx build \
  --platform linux/arm64 \
  -f Dockerfile.cluster \
  -t nielsrolf/ow-cluster:$VERSION \
  --load .

# Step 3: Build and push AMD64 to Docker Hub
docker buildx build \
  --platform linux/amd64 \
  -f Dockerfile.cluster \
  -t nielsrolf/ow-cluster:$VERSION \
  --push .

Note: The frontend must be built before the Docker image because the build process copies the pre-built static files from openweights/dashboard/backend/static/. This makes the Docker build much faster by avoiding Node.js installation and npm build steps.

Running cluster/dashboard image locally

The cluster image supports three modes via the OW_CMD environment variable:

  • cluster: Run only the cluster manager
  • serve: Run only the dashboard backend
  • both (default): Run both cluster manager and dashboard backend
# Get the current version
VERSION=$(python -c "from openweights.client.jobs import Jobs; print(Jobs.base_image.split(':')[-1])")

# Run cluster manager only
docker run --rm --env-file .env -e OW_CMD=cluster -ti nielsrolf/ow-cluster:$VERSION

# Run dashboard backend only
docker run --rm --env-file .env -e OW_CMD=serve -p 8124:8124 -ti nielsrolf/ow-cluster:$VERSION

# Run both (default)
docker run --rm --env-file .env -p 8124:8124 -ti nielsrolf/ow-cluster:$VERSION

# Interactive shell
docker run --rm --env-file .env -ti nielsrolf/ow-cluster:$VERSION /bin/bash

Dashboard environment variables

When running the dashboard backend, you can configure:

  • SITE_URL: The public URL where the dashboard is accessible (default: http://localhost:8124)
  • API_EXTERNAL_URL: The external API URL (default: http://localhost:8124)
  • ADDITIONAL_REDIRECT_URLS: Additional allowed redirect URLs for OAuth